Comparison of predictor sets for species richness and the number of rare species of butterflies and birds

Aim
Accurate inventories of biota are typically restricted to few locations within an extensive region. Accordingly, effective planning must involve some form of surrogate measures coupled with spatial modelling. We conducted a simultaneous comparison of models of both species richness and the number of rare species using three types of surrogates (indicator species, vegetation composition and structure, and topoclimate) as predictors. We evaluated each type of surrogate alone and in combination with others.

Location
Data for our analyses were collected from 1996–2004 in three adjacent mountain ranges in the central Great Basin (Lander and Nye counties, Nevada, USA), the Shoshone Mountains, Toiyabe Range and Toquima Range.

Methods
Data on species richness and species composition of butterflies and birds and measures of vegetation composition and structure were obtained in the field. Topoclimatic variables were derived by GIS from digital sources and satellite images. We used Poisson regression with Bayesian model averaging to predict species richness and the number of rare species. We compared the expected prediction success of all models on the basis of internal and external validation trials.

Results
Same-taxon indicator species were the most accurate predictors of species richness and of the number of rare species of butterflies and birds. Cross-taxon indicator species and topoclimate variables were reasonably accurate predictors of species richness of butterflies and birds and of the number of rare butterfly species. Although vegetation variables were more effective for predicting species richness and number of rare species of birds than of butterflies, they were the least accurate predictors overall.

Main conclusions
Although indicator species may provide the most accurate predictions of species richness, their practical value, like any surrogate measure, depends greatly on ecological considerations and land-use context. In general, the ability to predict numbers of rare species based on any set of candidate predictors was weaker than the ability to predict species richness, which may result from the high degree of stochasticity that often characterizes distributions of rare species. Our statistical approach for objective examination of different candidate predictors can help ensure that selection of species-richness surrogates in any system is scientifically reliable and cost-effective.